CN114881330A - Neural network-based rail transit passenger flow prediction method and system - Google Patents

Neural network-based rail transit passenger flow prediction method and system Download PDF

Info

Publication number
CN114881330A
CN114881330A CN202210498705.5A CN202210498705A CN114881330A CN 114881330 A CN114881330 A CN 114881330A CN 202210498705 A CN202210498705 A CN 202210498705A CN 114881330 A CN114881330 A CN 114881330A
Authority
CN
China
Prior art keywords
passenger flow
extraction channel
rail transit
data set
time sequence
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202210498705.5A
Other languages
Chinese (zh)
Other versions
CN114881330B (en
Inventor
王晨曦
张惠臻
潘玉彪
缑锦
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huaqiao University
Original Assignee
Huaqiao University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Huaqiao University filed Critical Huaqiao University
Priority to CN202210498705.5A priority Critical patent/CN114881330B/en
Publication of CN114881330A publication Critical patent/CN114881330A/en
Application granted granted Critical
Publication of CN114881330B publication Critical patent/CN114881330B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/40—Business processes related to the transportation industry
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Marketing (AREA)
  • Development Economics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Databases & Information Systems (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Multimedia (AREA)
  • Medical Informatics (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a rail transit passenger flow prediction method and a rail transit passenger flow prediction system based on a neural network, and belongs to the technical field of intelligent traffic, wherein the method comprises the following steps: s10, acquiring a card swiping data set of rail transit and an adjacency relation graph of rail stations, and preprocessing the card swiping data set to obtain a station entering passenger flow data set; step S20, reconstructing the site adjacency relation based on the adjacency relation graph and the inbound passenger flow data set; s30, constructing a consistent feature extraction channel and a heterogeneous feature extraction channel in a multidimensional space state based on GCN, and adaptively distributing weights of features extracted by the consistent feature extraction channel and the heterogeneous feature extraction channel through an attention i on mechanism; s40, constructing a time sequence prediction model based on the LSTM and the full-connection layer network, and training the time sequence prediction model; and step S50, inputting the station adjacency relation into a time sequence prediction model, and outputting a rail transit passenger flow prediction result. The invention has the advantages that: the accuracy of rail transit passenger flow prediction is greatly improved.

Description

Neural network-based rail transit passenger flow prediction method and system
Technical Field
The invention relates to the technical field of intelligent traffic, in particular to a rail transit passenger flow prediction method and a rail transit passenger flow prediction system based on a neural network.
Background
With the enlargement of the urban scale, rail transit develops into a main public trip mode of medium and large-sized cities; in this context, urban rails are under tremendous passenger traffic pressure and security pressure. Therefore, real-time passenger flow prediction plays a key role in planning a connection line and sharing rail traffic pressure and the like for a decision maker, and the predicted real-time passenger flow can be timely informed to passengers through a visual interface so as to avoid peak periods and reduce potential safety hazards.
Aiming at the prediction of rail transit passenger flow, an ARIMA linear model, SVM shallow machine learning, CNN or LSTM and other basic deep learning models are mainly adopted in the prior art; however, these models do not sufficiently recognize and extract nonlinear spatio-temporal features in the data, resulting in low prediction accuracy of the models.
Therefore, how to provide a method and a system for predicting rail transit passenger flow based on a neural network to improve the accuracy of rail transit passenger flow prediction becomes a technical problem to be solved urgently.
Disclosure of Invention
The invention aims to provide a method and a system for predicting rail transit passenger flow based on a neural network, so as to improve the accuracy of rail transit passenger flow prediction.
In a first aspect, the invention provides a rail transit passenger flow prediction method based on a neural network, which comprises the following steps:
s10, acquiring a card swiping data set of rail transit and an adjacency relation graph of rail stations, and preprocessing the card swiping data set to obtain an inbound passenger flow data set;
step S20, reconstructing the station adjacency relation based on the adjacency relation graph and the inbound passenger flow data set;
s30, constructing a consistent feature extraction channel and a heterogeneous feature extraction channel in a multidimensional space state based on GCN, and adaptively distributing the weight of the features extracted by the consistent feature extraction channel and the heterogeneous feature extraction channel through an Attention mechanism;
step S40, constructing a time sequence prediction model through an LSTM and a full-connection layer network based on the consistency feature extraction channel and the heterogeneity feature extraction channel, and training the time sequence prediction model;
and step S50, inputting the station adjacency relation into the trained time sequence prediction model, and outputting a rail transit passenger flow prediction result.
Further, the step S10 is specifically:
acquiring a card swiping data set of the rail transit through a gate, and acquiring an adjacency relation graph of a rail station through map software;
and after the card swiping data set is cleaned, constructing a passenger flow matrix according to the site and time period of each card swiping data in the card swiping data set, and further generating an inbound passenger flow data set.
Further, the step S20 is specifically:
and reconstructing the site adjacency relation based on the adjacency relation matrix constructed by the adjacency relation graph and the passenger flow matrix.
Further, the step S30 is specifically:
constructing a heterogeneous characteristic extraction channel based on the adjacency matrix, and constructing a consistent characteristic extraction channel based on the passenger flow matrix;
and introducing a feature fusion module through an Attention mechanism, and further adaptively distributing the weight of the features extracted by the consistent feature extraction channel and the heterogeneous feature extraction channel.
Further, in step S40, the training of the time sequence prediction model specifically includes:
and training the time sequence prediction model by using a pre-established test set until the root mean square error and the average error between the predicted value and the true value of the time sequence prediction model are smaller than a set threshold value.
In a second aspect, the invention provides a rail transit passenger flow prediction system based on a neural network, which comprises the following modules:
the system comprises a data acquisition module, a passenger flow data acquisition module and a passenger flow data acquisition module, wherein the data acquisition module is used for acquiring a card swiping data set of the rail transit and an adjacency relation graph of a rail station, and preprocessing the card swiping data set to obtain a passenger flow data set entering the station;
the station adjacency relation reconstruction module is used for reconstructing station adjacency relation based on the adjacency relation graph and the inbound passenger flow data set;
the characteristic extraction channel construction module is used for constructing a consistent characteristic extraction channel and a heterogeneous characteristic extraction channel in a multi-dimensional space state based on GCN, and adaptively distributing the weight of the extracted characteristics of the consistent characteristic extraction channel and the heterogeneous characteristic extraction channel through an Attention mechanism;
the time sequence prediction model creating module is used for constructing a time sequence prediction model through an LSTM and a full-connection layer network based on the consistency characteristic extraction channel and the heterogeneity characteristic extraction channel and training the time sequence prediction model;
and the rail transit passenger flow prediction module is used for inputting the station adjacency relation into the trained time sequence prediction model and outputting a rail transit passenger flow prediction result.
Further, the data acquisition module specifically includes:
acquiring a card swiping data set of the rail transit through a gate, and acquiring an adjacency relation graph of a rail station through map software;
and after the card swiping data set is cleaned, constructing a passenger flow matrix according to the site and time period of each card swiping data in the card swiping data set, and further generating an inbound passenger flow data set.
Further, the site adjacency relationship reconstruction module specifically includes:
and reconstructing the site adjacency relation based on the adjacency relation matrix constructed by the adjacency relation graph and the passenger flow matrix.
Further, the feature extraction channel construction module specifically includes:
constructing a heterogeneous characteristic extraction channel based on the adjacency matrix, and constructing a consistent characteristic extraction channel based on the passenger flow matrix;
and introducing a feature fusion module through an Attention mechanism, and further adaptively distributing the weight of the features extracted by the consistent feature extraction channel and the heterogeneous feature extraction channel.
Further, in the time sequence prediction model creating module, the training of the time sequence prediction model specifically includes:
and training the time sequence prediction model by using a pre-established test set until the root mean square error and the average error between the predicted value and the true value of the time sequence prediction model are smaller than a set threshold value.
The invention has the advantages that:
the method comprises the steps of preprocessing a card swiping data set to obtain an inbound passenger flow data set by obtaining a card swiping data set and an adjacency relation graph, reconstructing a site adjacency relation based on the adjacency relation graph and the inbound passenger flow data set, fully mining hidden passenger flow moving trends in the card swiping data set compared with the traditional predefined node adjacency relation, reconstructing the site adjacency relation according to the trends, and therefore accurately mining the adjacency relation of track sites; constructing a consistent feature extraction channel and a heterogeneous feature extraction channel in a multi-dimensional space state through the GCN, namely introducing independent parameters and shared parameters to fully excavate spatial correlation in the multi-dimensional space, and correcting errors of the GCN based on predefined adjacency relations; the characteristics are adaptively fused through an Attention mechanism, errors caused by self-defined weight are reduced, and finally the accuracy of rail transit passenger flow prediction is greatly improved.
Drawings
The invention will be further described with reference to the following examples with reference to the accompanying drawings.
Fig. 1 is a flow chart of a method for predicting rail transit passenger flow based on a neural network according to the present invention.
Fig. 2 is a schematic structural diagram of a rail transit passenger flow prediction system based on a neural network.
FIG. 3 is a schematic diagram of a time sequence prediction model according to the present invention.
Fig. 4 is a schematic diagram of the heterogeneous feature extraction channel of the present invention.
FIG. 5 is a schematic diagram of the consistent feature extraction channel of the invention.
FIG. 6 is a schematic representation of the LSTM of the present invention.
FIG. 7 is a schematic diagram of the Attention mechanism of the present invention.
Detailed Description
Referring to fig. 1 to 7, the technical solution in the embodiment of the present application has the following general idea: acquiring a card swiping data set, preprocessing the card swiping data set to obtain an inbound passenger flow data set, and reconstructing a site adjacency relation according to an adjacency relation graph of track sites and the inbound passenger flow data set to obtain a site adjacency graph in a multi-dimensional state; correcting the GCN model based on the multidimensional site adjacency relation, and establishing a consistency feature extraction channel and a heterogeneity feature extraction channel to respectively extract consistency and heterogeneity in a multidimensional space; establishing a time sequence prediction model based on the dual-channel GCN module, the Attention module, the LSTM model and the full connection layer, training the time sequence prediction model until convergence, and predicting passenger flow by the trained time sequence prediction model; the accuracy of rail transit passenger flow prediction is improved by fully considering the adjacency relation among rail stations, a decision maker can be helped to timely cope with the peak time of sudden passenger flow, security force is timely deployed, and safety risks are avoided.
Example one
The invention discloses a better embodiment of a rail transit passenger flow prediction method based on a neural network, which comprises the following steps:
s10, acquiring a card swiping data set of rail transit and an adjacency relation graph of rail stations, and preprocessing the card swiping data set to obtain an inbound passenger flow data set;
step S20, reconstructing the site adjacency relation based on the adjacency relation graph and the inbound passenger flow data set, namely obtaining the site adjacency graph in a multi-dimensional state;
step S30, constructing a consistent feature extraction channel and a heterogeneous feature extraction channel in a multidimensional space state based on GCN (graph convolution network), and adaptively distributing the weight of the features extracted by the consistent feature extraction channel and the heterogeneous feature extraction channel through an Attention mechanism;
step S40, constructing a time sequence prediction model through an LSTM and a full-connection layer network based on the consistency feature extraction channel and the heterogeneity feature extraction channel, and training the time sequence prediction model;
and step S50, inputting the station adjacency relation into the trained time sequence prediction model, and outputting a rail transit passenger flow prediction result.
The step S10 specifically includes:
acquiring a card swiping data set of the rail transit through a gate, and acquiring an adjacency relation graph of a rail station through map software;
after the card swiping data set is cleaned, a passenger flow matrix is constructed according to the station and the time period (the time periods are divided according to different prediction granularities) where each piece of card swiping data in the card swiping data set is located, and then the inbound passenger flow data set is generated.
The step S20 specifically includes:
and reconstructing the site adjacency relation based on the adjacency relation matrix constructed by the adjacency relation graph and the passenger flow matrix.
The adjacency matrix A T The formula of (1) is as follows:
Figure BDA0003633927330000061
the passenger flow matrix A F The formula of (1) is as follows:
Figure BDA0003633927330000062
the step S30 specifically includes:
constructing a heterogeneous characteristic extraction channel based on the adjacent matrix and constructing a consistent characteristic extraction channel based on the passenger flow matrix by utilizing the DGCN and the CGCN;
respectively constructing a DGCN module and a CGCN module by combining adjacent matrixes of track stations in a physical space and a non-physical space and passenger flow matrix data, wherein the DGCN module constructs a heterogeneous characteristic extraction channel based on the idea of independent parameters, and the CGCN constructs a consistent characteristic extraction channel based on the idea of shared parameters;
extracting indirect adjacent characteristics between nodes by using a layer I GCN network, and obtaining space correlation characteristics Z under a multidimensional space based on independent network parameters D And Z C The GCN extraction rule is as follows:
Figure BDA0003633927330000063
construction of multilayer GCN network based on same parameter theta for extracting consistency information Z under two spaces CT And Z CF ：
Figure BDA0003633927330000064
Figure BDA0003633927330000065
And introducing a feature fusion module through an Attention mechanism, and further adaptively distributing the weight of the features extracted by the consistent feature extraction channel and the heterogeneous feature extraction channel.
The feature fusion layer of the Attention mechanism comprises an Attention score calculation module and an Attention fusion block; the attention score calculating module calculates a correlation coefficient between the output of each time step of the encoder and the output of the encoder, and the attention fusion module carries out weighted summation based on the scores of all parts; therefore, the acquired features are adaptively fused through the process to obtain fused features:
Figure BDA0003633927330000071
Z＝α 1 ·Z D +α 2 ·Z C +α 3 ·Z CT +α 4 ·Z CF 。
in step S40, the training of the time sequence prediction model specifically includes:
training the time sequence prediction model by using a pre-established test set until the root mean square error and the average error between the predicted value and the true value of the time sequence prediction model are smaller than a set threshold;
the root mean square error is calculated as follows:
Figure BDA0003633927330000072
the calculation formula of the average error is as follows:
Figure BDA0003633927330000073
wherein RMSE represents the root mean square error; MAE represents mean error; n represents the data volume of the test set; y is i Represents the true value;
Figure BDA0003633927330000074
representing a predicted value; i is a positive integer, and i is more than or equal to 1 and less than or equal to n.
In step S40, the time-series prediction model is a prediction output layer based on LSTM and the full-link layer, and the part first performs time-series correlation extraction on the fusion feature Z using LSTM to obtain an output { hh 1 ,h 2 ,…,h t And the part is used as the input learning distributed feature of a full connection layer to represent the mapping of the input learning distributed feature to a sample space, and finally the predicted value of the track passenger flow at the moment of t +1 is obtained
Figure BDA0003633927330000075
Example two
The invention discloses a preferable embodiment of a rail transit passenger flow prediction system based on a neural network, which comprises the following modules:
the system comprises a data acquisition module, a passenger flow data acquisition module and a passenger flow data acquisition module, wherein the data acquisition module is used for acquiring a card swiping data set of the rail transit and an adjacency relation graph of a rail station, and preprocessing the card swiping data set to obtain a passenger flow data set entering the station;
the station adjacency relation reconstruction module is used for reconstructing the station adjacency relation based on the adjacency relation graph and the inbound passenger flow data set, so that the station adjacency graph in the multi-dimensional state is obtained;
the characteristic extraction channel construction module is used for constructing a consistent characteristic extraction channel and a heterogeneous characteristic extraction channel in a multidimensional space state based on GCN (graph convolution network), and adaptively distributing the weight of the extracted characteristics of the consistent characteristic extraction channel and the heterogeneous characteristic extraction channel through an Attention mechanism;
the time sequence prediction model creating module is used for constructing a time sequence prediction model through an LSTM and a full-connection layer network based on the consistency characteristic extraction channel and the heterogeneity characteristic extraction channel and training the time sequence prediction model;
and the rail transit passenger flow prediction module is used for inputting the station adjacency relation into the trained time sequence prediction model and outputting a rail transit passenger flow prediction result.
The data acquisition module specifically comprises:
acquiring a card swiping data set of the rail transit through a gate, and acquiring an adjacency relation graph of a rail station through map software;
after the card swiping data set is cleaned, a passenger flow matrix is constructed according to the station and the time period (the time periods are divided according to different prediction granularities) where each piece of card swiping data in the card swiping data set is located, and then the inbound passenger flow data set is generated.
The site adjacency relationship reconstruction module specifically comprises:
and reconstructing the site adjacency relation based on the adjacency relation matrix constructed by the adjacency relation graph and the passenger flow matrix.
The adjacency matrix A T The formula of (1) is as follows:
Figure BDA0003633927330000081
the passenger flow matrix A F The formula of (1) is as follows:
Figure BDA0003633927330000091
the feature extraction channel construction module specifically comprises:
constructing a heterogeneous characteristic extraction channel based on the adjacent matrix and constructing a consistent characteristic extraction channel based on the passenger flow matrix by utilizing the DGCN and the CGCN;
respectively constructing a DGCN module and a CGCN module by combining adjacent matrixes of track stations in a physical space and a non-physical space and passenger flow matrix data, wherein the DGCN module constructs a heterogeneous characteristic extraction channel based on the idea of independent parameters, and the CGCN constructs a consistent characteristic extraction channel based on the idea of shared parameters;
extracting indirect adjacent characteristics between nodes by using a layer I GCN network, and obtaining space correlation characteristics Z under a multidimensional space based on independent network parameters D And Z C The GCN extraction rule is as follows:
Figure BDA0003633927330000092
construction of multilayer GCN network based on same parameter theta for extracting consistency information Z under two spaces CT And Z CF ：
Figure BDA0003633927330000093
Figure BDA0003633927330000094
And introducing a feature fusion module through an Attention mechanism, and further adaptively distributing the weight of the features extracted by the consistent feature extraction channel and the heterogeneous feature extraction channel.
The feature fusion layer of the Attention mechanism comprises an Attention score calculation module and an Attention fusion block; the attention score calculating module calculates a correlation coefficient between the output of each time step of the encoder and the output of the encoder, and the attention fusion module carries out weighted summation based on the scores of all parts; therefore, the acquired features are adaptively fused through the process to obtain fused features:
Figure BDA0003633927330000095
Z＝α 1 ·Z D +α 2 ·Z C +α 3 ·Z CT +α 4 ·Z CF 。
in the time sequence prediction model creating module, the training of the time sequence prediction model specifically includes:
training the time sequence prediction model by using a pre-established test set until the root mean square error and the average error between the predicted value and the true value of the time sequence prediction model are smaller than a set threshold;
the root mean square error is calculated as follows:
Figure BDA0003633927330000101
the calculation formula of the average error is as follows:
Figure BDA0003633927330000102
wherein RMSE represents the root mean square error; MAE represents mean error; n represents the data volume of the test set; y is i Represents the true value;
Figure BDA0003633927330000103
representing a predicted value; i is a positive integer, and i is more than or equal to 1 and less than or equal to n.
In step S40, the time-series prediction model is a prediction output layer based on LSTM and the full-link layer, and the time-series correlation extraction is performed on the fusion feature Z by using LSTM to obtain an output { h |, where 1 ,h 2 ,...,h t And the part is used as the input learning distributed feature of a full connection layer to represent the mapping of the input learning distributed feature to a sample space, and finally the predicted value of the track passenger flow at the moment of t +1 is obtained
Figure BDA0003633927330000104
EXAMPLE III
Acquiring a CSV-format card swiping data set of rail transit:
field(s) Name of field Field description Data examples
Card_ID Number for punching card Recording unique value of serial number of card used by passenger 6.0214**28
Date Date of transaction The date of transaction is recorded 2015/4/17
Time Transaction time The record transaction time 13:49:58
Station_Name Transaction site Site where transaction is located Number 4 line shen zhuang station
Type Ride type Recording ride type of public transportation Subway
Distance Mileage Recording the distance of ride 4.5
Attribute Type (B) Recording whether to enjoy discount Non-preferential
Preprocessing the card swiping data set, and counting inbound passenger flow data of different stations in different time periods according to the station and time period (10min) of each card swiping data in the card swiping data set:
Figure BDA0003633927330000111
building a time sequence prediction model of short-time passenger flow of the track based on a dual-channel GCN model, and analyzing time complexity and space complexity respectively by using two indexes of FLOPs and Params; wherein, the time complexity is determined by the operation times FLOPs (0.0175G) of the model, and the space complexity is determined by the number of parameters Params (2982574); the parameter settings of the time series prediction model are shown in the following table:
parameter(s) Content of model selection
Training set: test set 3:01
Activating a function tanh
Learning rate 0.0001
Optimizer Adam
Loss function MSE
Step of time 5
Number of iteration rounds 500
The time sequence prediction model is verified, the RMSE and the MAE can better embody the error between the true value and the predicted value, so the RMSE and the MAE are adopted to compare the proposed time sequence prediction model with a basic mathematical statistic model (HA \ ARIMA), a machine learning model (SVR) and a deep learning model LSTM, and the experimental results are shown in the following table:
Figure BDA0003633927330000112
Figure BDA0003633927330000121
the experimental result shows that the time sequence prediction model can obtain a more accurate prediction result, and is beneficial to providing a favorable basis for making a subsequent decision.
In summary, the invention has the advantages that:
the method comprises the steps of preprocessing a card swiping data set to obtain an inbound passenger flow data set by obtaining a card swiping data set and an adjacency relation graph, reconstructing a site adjacency relation based on the adjacency relation graph and the inbound passenger flow data set, fully mining hidden passenger flow moving trends in the card swiping data set compared with the traditional predefined node adjacency relation, reconstructing the site adjacency relation according to the trends, and therefore accurately mining the adjacency relation of track sites; constructing a consistent feature extraction channel and a heterogeneous feature extraction channel in a multi-dimensional space state through the GCN, namely introducing independent parameters and shared parameters to fully excavate spatial correlation in the multi-dimensional space, and correcting errors of the GCN based on predefined adjacency relations; the characteristics are adaptively fused through an Attention mechanism, errors caused by self-defined weight are reduced, and finally the accuracy of rail transit passenger flow prediction is greatly improved.
Although specific embodiments of the invention have been described above, it will be understood by those skilled in the art that the specific embodiments described are illustrative only and are not limiting upon the scope of the invention, and that equivalent modifications and variations can be made by those skilled in the art without departing from the spirit of the invention, which is to be limited only by the appended claims.

Claims (10)

1. A rail transit passenger flow prediction method based on a neural network is characterized by comprising the following steps: the method comprises the following steps:
s10, acquiring a card swiping data set of rail transit and an adjacency relation graph of rail stations, and preprocessing the card swiping data set to obtain an inbound passenger flow data set;
step S20, reconstructing the station adjacency relation based on the adjacency relation graph and the inbound passenger flow data set;
s30, constructing a consistent feature extraction channel and a heterogeneous feature extraction channel in a multidimensional space state based on GCN, and adaptively distributing the weight of the features extracted by the consistent feature extraction channel and the heterogeneous feature extraction channel through an Attention mechanism;
step S40, constructing a time sequence prediction model through an LSTM and a full-connection layer network based on the consistency feature extraction channel and the heterogeneity feature extraction channel, and training the time sequence prediction model;
and step S50, inputting the station adjacency relation into the trained time sequence prediction model, and outputting a rail transit passenger flow prediction result.
2. The method for predicting the passenger flow of the rail transit based on the neural network as claimed in claim 1, wherein: the step S10 specifically includes:
acquiring a card swiping data set of the rail transit through a gate, and acquiring an adjacency relation graph of a rail station through map software;
and after the card swiping data set is cleaned, constructing a passenger flow matrix according to the site and time period of each card swiping data in the card swiping data set, and further generating an inbound passenger flow data set.
3. The method for predicting the passenger flow of the rail transit based on the neural network as claimed in claim 2, wherein: the step S20 specifically includes:
and reconstructing the site adjacency relation based on the adjacency relation matrix constructed by the adjacency relation graph and the passenger flow matrix.
4. The method for predicting rail transit passenger flow based on the neural network as claimed in claim 3, wherein: the step S30 specifically includes:
constructing a heterogeneous characteristic extraction channel based on the adjacency matrix, and constructing a consistent characteristic extraction channel based on the passenger flow matrix;
and introducing a feature fusion module through an Attention mechanism, and further adaptively distributing the weight of the features extracted by the consistent feature extraction channel and the heterogeneous feature extraction channel.
5. The method for predicting the passenger flow of the rail transit based on the neural network as claimed in claim 1, wherein: in step S40, the training of the time sequence prediction model specifically includes:
and training the time sequence prediction model by using a pre-established test set until the root mean square error and the average error between the predicted value and the true value of the time sequence prediction model are smaller than a set threshold value.
6. A rail transit passenger flow prediction system based on a neural network is characterized in that: the system comprises the following modules:
the system comprises a data acquisition module, a passenger flow data acquisition module and a passenger flow data acquisition module, wherein the data acquisition module is used for acquiring a card swiping data set of the rail transit and an adjacency relation graph of a rail station, and preprocessing the card swiping data set to obtain a passenger flow data set entering the station;
the station adjacency relation reconstruction module is used for reconstructing station adjacency relation based on the adjacency relation graph and the inbound passenger flow data set;
the characteristic extraction channel construction module is used for constructing a consistent characteristic extraction channel and a heterogeneous characteristic extraction channel in a multi-dimensional space state based on GCN, and adaptively distributing the weight of the extracted characteristics of the consistent characteristic extraction channel and the heterogeneous characteristic extraction channel through an Attention mechanism;
the time sequence prediction model creating module is used for constructing a time sequence prediction model through an LSTM and a full-connection layer network based on the consistency characteristic extraction channel and the heterogeneity characteristic extraction channel and training the time sequence prediction model;
and the rail transit passenger flow prediction module is used for inputting the station adjacency relation into the trained time sequence prediction model and outputting a rail transit passenger flow prediction result.
7. The neural network-based rail transit passenger flow prediction system of claim 6, wherein: the data acquisition module specifically comprises:
acquiring a card swiping data set of the rail transit through a gate, and acquiring an adjacency relation graph of a rail station through map software;
and after the card swiping data set is cleaned, constructing a passenger flow matrix according to the site and time period of each card swiping data in the card swiping data set, and further generating an inbound passenger flow data set.
8. The neural network-based rail transit passenger flow prediction system of claim 7, wherein: the site adjacency relationship reconstruction module specifically comprises:
and reconstructing the site adjacency relation based on the adjacency relation matrix constructed by the adjacency relation graph and the passenger flow matrix.
9. The neural network-based rail transit passenger flow prediction system of claim 8, wherein: the feature extraction channel construction module specifically comprises:
constructing a heterogeneous characteristic extraction channel based on the adjacency matrix, and constructing a consistent characteristic extraction channel based on the passenger flow matrix;
and introducing a feature fusion module through an Attention mechanism, and further adaptively distributing the weight of the features extracted by the consistent feature extraction channel and the heterogeneous feature extraction channel.
10. The neural network-based rail transit passenger flow prediction system of claim 6, wherein: in the time sequence prediction model creating module, the training of the time sequence prediction model specifically includes:
and training the time sequence prediction model by using a pre-established test set until the root mean square error and the average error between the predicted value and the true value of the time sequence prediction model are smaller than a set threshold value.
CN202210498705.5A 2022-05-09 2022-05-09 Rail transit passenger flow prediction method and system based on neural network Active CN114881330B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210498705.5A CN114881330B (en) 2022-05-09 2022-05-09 Rail transit passenger flow prediction method and system based on neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210498705.5A CN114881330B (en) 2022-05-09 2022-05-09 Rail transit passenger flow prediction method and system based on neural network

Publications (2)

Publication Number Publication Date
CN114881330A true CN114881330A (en) 2022-08-09
CN114881330B CN114881330B (en) 2024-08-06

Family

ID=82673721

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210498705.5A Active CN114881330B (en) 2022-05-09 2022-05-09 Rail transit passenger flow prediction method and system based on neural network

Country Status (1)

Country Link
CN (1) CN114881330B (en)

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108537392A (en) * 2018-04-26 2018-09-14 重庆大学 A method of standee's flow is pointed out based on space-time characteristic predicted orbit traffic station
CN110222873A (en) * 2019-05-14 2019-09-10 重庆邮电大学 A kind of subway station passenger flow forecast method based on big data
CN110942411A (en) * 2019-11-20 2020-03-31 北京城建设计发展集团股份有限公司 Subway station passenger flow grading alarm method based on regional linkage
CN111667092A (en) * 2020-04-21 2020-09-15 北京交通大学 Method and system for predicting short-term passenger flow of rail transit based on graph convolution neural network
CN111738535A (en) * 2020-08-25 2020-10-02 北京交通大学 Method, device, equipment and storage medium for predicting rail transit time-space short-time passenger flow
CN111931978A (en) * 2020-06-29 2020-11-13 南京熊猫电子股份有限公司 Urban rail transit passenger flow state prediction method based on space-time characteristics
CN112001548A (en) * 2020-08-25 2020-11-27 北京交通大学 OD passenger flow prediction method based on deep learning
CN112418547A (en) * 2020-12-03 2021-02-26 北京工业大学 Bus stop passenger flow prediction method based on GCN-LSTM combined model
US20210209939A1 (en) * 2020-12-08 2021-07-08 Harbin Engineering University Large-scale real-time traffic flow prediction method based on fuzzy logic and deep LSTM
CN113177657A (en) * 2021-04-20 2021-07-27 上海大学 Rail transit passenger flow prediction method and device
CN114399099A (en) * 2021-12-30 2022-04-26 中信云网有限公司 Method and device for predicting passenger flow of urban rail transit section

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108537392A (en) * 2018-04-26 2018-09-14 重庆大学 A method of standee's flow is pointed out based on space-time characteristic predicted orbit traffic station
CN110222873A (en) * 2019-05-14 2019-09-10 重庆邮电大学 A kind of subway station passenger flow forecast method based on big data
CN110942411A (en) * 2019-11-20 2020-03-31 北京城建设计发展集团股份有限公司 Subway station passenger flow grading alarm method based on regional linkage
CN111667092A (en) * 2020-04-21 2020-09-15 北京交通大学 Method and system for predicting short-term passenger flow of rail transit based on graph convolution neural network
CN111931978A (en) * 2020-06-29 2020-11-13 南京熊猫电子股份有限公司 Urban rail transit passenger flow state prediction method based on space-time characteristics
CN111738535A (en) * 2020-08-25 2020-10-02 北京交通大学 Method, device, equipment and storage medium for predicting rail transit time-space short-time passenger flow
CN112001548A (en) * 2020-08-25 2020-11-27 北京交通大学 OD passenger flow prediction method based on deep learning
CN112418547A (en) * 2020-12-03 2021-02-26 北京工业大学 Bus stop passenger flow prediction method based on GCN-LSTM combined model
US20210209939A1 (en) * 2020-12-08 2021-07-08 Harbin Engineering University Large-scale real-time traffic flow prediction method based on fuzzy logic and deep LSTM
CN113177657A (en) * 2021-04-20 2021-07-27 上海大学 Rail transit passenger flow prediction method and device
CN114399099A (en) * 2021-12-30 2022-04-26 中信云网有限公司 Method and device for predicting passenger flow of urban rail transit section

Also Published As

Publication number Publication date
CN114881330B (en) 2024-08-06

Similar Documents

Publication Publication Date Title
CN111540193B (en) Traffic data restoration method for generating countermeasure network based on graph convolution time sequence
CN110263111B (en) Priori knowledge-based space-time monitoring method for land utilization/coverage information
CN111079977B (en) Heterogeneous federal learning mine electromagnetic radiation trend tracking method based on SVD algorithm
CN111709244B (en) Deep learning method for identifying cause and effect relationship of contradictory dispute
CN107529651A (en) A kind of urban transportation passenger flow forecasting and equipment based on deep learning
CN102508907B (en) Dynamic recommendation method based on training set optimization for recommendation system
CN108446794A (en) One kind being based on multiple convolutional neural networks combination framework deep learning prediction techniques
CN110515931B (en) Capacitive type equipment defect prediction method based on random forest algorithm
CN112054943B (en) Traffic prediction method for mobile network base station
CN103995237A (en) Satellite power supply system online fault diagnosis method
CN113988348B (en) Air quality prediction method for grid monitoring
CN111144281B (en) Urban rail transit OD passenger flow estimation method based on machine learning
CN113327417B (en) Traffic flow prediction method based on 3D dynamic space-time residual convolution associated network
CN115827335B (en) Time sequence data missing interpolation system and time sequence data missing interpolation method based on modal crossing method
CN109034192B (en) Track-vehicle body vibration state prediction method based on deep learning
CN101964061B (en) Binary kernel function support vector machine-based vehicle type recognition method
CN117093830A (en) User load data restoration method considering local and global
CN110716998B (en) Fine scale population data spatialization method
CN116400433A (en) Single-station time-by-time air temperature forecasting method based on data fusion and mixed convolution
CN114881330A (en) Neural network-based rail transit passenger flow prediction method and system
CN116541593B (en) Course recommendation method based on hypergraph neural network
CN109472115B (en) Large-scale complex network modeling method and device based on geographic information
CN107220742A (en) A kind of development of information system common support method analyzed based on system vulnerability and platform
CN117371571A (en) Regional air quality prediction model based on multi-scale dynamic synchronous diagram mechanism
CN115905434B (en) Road network track completion method based on learning interpolation prediction

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant